I have been running a small research lab that is working on ways to increase intelligence in machine learning systems. Interested in funding our work? Reach out via [email protected].
State of the art on ARC-AGI
The Abstraction and Reasoning Corpus (ARC) is one of the most well-respected general artificial intelligence benchmarks. It has proven incredibly difficult for the AI/ML world. Foundation models such as GPT-4 have not faired well at the challenge. Several world-renowned labs and thousands of individuals have made attempts without success. Utilizing a relentless empirical process, we were able to surpass all previous attempts and reach state of the art performance and are currently ranking #1 on the leaderboard of ARC-Prize with a world-record score of 46%. Our team consists of Jack Cole and Mohamed Osman (1st place ARCathon 2023) and Michael Hodel (1st place ARCathon 2022). To learn more about our approach, see this post or Dr. Tim Scarfe’s interview with us on the Machine Learning Street Talk.